📊 Linear Discriminant Analysis (LDA)

Dimensionality reduction and multi-class classification on the UCI Wine dataset using Linear Discriminant Analysis.

What It Does

Reduces 13 chemical features of wine samples down to 2 linear discriminants using LDA, then trains a classifier to predict customer segments — with decision-boundary visualizations for both training and test sets.

Methodology

1. Load the Wine dataset (178 samples, 13 features, 3 classes)
2. 80/20 train/test split + standard feature scaling
3. Apply LDA → project onto 2 discriminant components
4. Train a classifier (Logistic Regression in Python, SVM in R)
5. Evaluate with a confusion matrix + accuracy score
6. Plot 2D decision boundaries for train and test sets

Dataset

Wine.csv — 178 samples across 3 customer segments, with 13 chemical analysis features (Alcohol, Malic Acid, Ash, Magnesium, Phenols, Flavanoids, etc.). Based on the UCI Wine dataset.

Quick Start

Python

pip install numpy matplotlib pandas scikit-learn
python lda.py

R

install.packages(c("caTools", "MASS", "e1071", "ElemStatLearn"))
source("lda.R")

🛠 Tech Stack

	Tool	Purpose
🐍	scikit-learn	LDA, Logistic Regression, StandardScaler
📊	matplotlib	Decision-boundary visualization
🧮	pandas / numpy	Data loading and manipulation
📈	MASS (R)	LDA implementation
🤖	e1071 (R)	SVM classifier

⚠️ Known Issues

• R: ElemStatLearn removed from CRAN — The ElemStatLearn package used for visualization in lda.R has been archived. Install from the CRAN archive or use an alternative plotting approach.

License

MIT — see LICENSE

Author

Kaustabh Ganguly (@stabgan)